3D Perception
Point Cloud Processing
import open3d as o3d
pcd = o3d.io.read_point_cloud("scene.ply")
pcd_down = pcd.voxel_down_sample(voxel_size=0.02)
pcd_down.estimate_normals(search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.1, max_nn=30))
Core Tasks
| Task |
Methods |
Libraries |
| Depth estimation |
MiDaS, Depth Anything, stereo |
torch, timm |
| 3D reconstruction |
NeRF, 3DGS, TSDF fusion |
nerfstudio, Open3D |
| 6DoF pose |
FoundationPose, MegaPose |
PyTorch3D |
| 3D segmentation |
MinkowskiEngine, PTv3 |
MinkowskiEngine |
| Scene graphs |
3DSSG, ConceptGraphs |
custom |
| Registration |
ICP, GICP, RANSAC |
Open3D, kiss-icp |
Coordinate Conventions
- ROS: X-forward, Y-left, Z-up (REP-103)
- OpenCV: X-right, Y-down, Z-forward
- OpenGL: X-right, Y-up, Z-backward
- Always document which convention is used
Key Libraries
Open3D, PyTorch3D, kaolin, MinkowskiEngine, nerfstudio
1---2name: 3d-perception3description: 3D vision pipelines. Point clouds, depth estimation, 3D reconstruction, pose estimation, scene understanding.4---56# 3D Perception78## Point Cloud Processing9```python10import open3d as o3d11pcd = o3d.io.read_point_cloud("scene.ply")12pcd_down = pcd.voxel_down_sample(voxel_size=0.02)13pcd_down.estimate_normals(search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.1, max_nn=30))14```1516## Core Tasks17| Task | Methods | Libraries |18|------|---------|-----------|19| Depth estimation | MiDaS, Depth Anything, stereo | torch, timm |20| 3D reconstruction | NeRF, 3DGS, TSDF fusion | nerfstudio, Open3D |21| 6DoF pose | FoundationPose, MegaPose | PyTorch3D |22| 3D segmentation | MinkowskiEngine, PTv3 | MinkowskiEngine |23| Scene graphs | 3DSSG, ConceptGraphs | custom |24| Registration | ICP, GICP, RANSAC | Open3D, kiss-icp |2526## Coordinate Conventions27- ROS: X-forward, Y-left, Z-up (REP-103)28- OpenCV: X-right, Y-down, Z-forward29- OpenGL: X-right, Y-up, Z-backward30- Always document which convention is used3132## Key Libraries33Open3D, PyTorch3D, kaolin, MinkowskiEngine, nerfstudio
aselimc